Model comparison
Compare Voyage Rerank 3 and Jina Reranker v3.5 using the same provider-sourced reranking & retrieval rubric. No mystery score and no invented benchmark ranking.
Facts checked September 4, 2026
Quick take
Voyage AI's quality-focused preview reranker, billed by all query and candidate tokens processed in each request.
Treat it as a preview and confirm candidate caps, language behavior, retention, regional processing, and production support directly before launch.
Jina's compact listwise reranker for long, multilingual, structured, legal, financial, and domain-specific candidate sets.
The open-weight license is non-commercial, not Apache or MIT. Provider latency figures use specific A100 test setups and should not be treated as your production SLA.
Compare the published facts
Values use each provider's own published units and limits. A blank means the provider did not publish a directly comparable value in the sources reviewed.
| Reranking & retrieval | Voyage Rerank 3 | Jina Reranker v3.5 |
|---|---|---|
| Price basisThe provider's billing unit, with the document or token assumptions needed to compare it fairly. | $0.05 / 1M processed tokens | Token packages; no single public dollar rate |
| Text capacityHow much query and candidate text the model can consider in one scoring pass. | 32,000 tokens | 131K tokens shared by query and candidates |
| Candidates per requestThe maximum number of possible results accepted in one request, when the provider publishes it. | Not published for the Rerank 3 preview | No fixed count published; bounded by shared context |
| Language coveragePublished multilingual support; an exact count is shown only when the provider gives one. | Not yet detailed for the preview | 93 supported; trained on 24 |
| Structured dataWhether the model is documented for JSON, tables, XML, or other data beyond normal prose. | Text candidates; structured-data support not stated | Tables, JSON, XML, legal and domain text |
| Where it runsHosted API, cloud marketplace, private deployment, or self-hosted weights. | Voyage hosted API and SDKs | Jina API, Elastic, clouds, Hugging Face, air-gapped |
How to choose
Start with the job you need to complete, then validate cost, access, and policy details on your exact provider route.
The reviewed Rerank 3 and pricing documentation does not publish a complete prompt-retention or model-training commitment. Obtain current data-processing terms before sending private retrieval candidates.
Provider and API links
Self-hosted and air-gapped deployments keep request data inside infrastructure you control. The reviewed hosted API pages do not state a complete retention or training-use commitment, so confirm Jina or Elastic terms for sensitive search data.
Frequently asked questions
Voyage Rerank 3: Voyage AI's quality-focused preview reranker, billed by all query and candidate tokens processed in each request. Jina Reranker v3.5: Jina's compact listwise reranker for long, multilingual, structured, legal, financial, and domain-specific candidate sets.
Consider Voyage Rerank 3 when your priority is Teams already using Voyage embeddings. Consider Jina Reranker v3.5 when your priority is Long candidate lists that benefit from listwise comparison. Test both with your own data and provider route before committing.
No. This comparison aligns provider-published facts for the Reranking & retrieval category. It does not claim a universal winner or combine incompatible third-party benchmark scores.
